What it actually shows you
Built from real usage pain, not a features checklist.
Rate limit gauges
Session (5hr) and weekly usage, with reset countdowns — the same numbers Claude Code's own /usage shows, always visible, updating live.
Spike-sensitive burn rate
A rolling 5-minute burn rate compared against your own trailing baseline — flags a runaway session while it's happening, not after you've already burned through your window.
Agent spawn accounting
Every subagent spawn, grouped by the request that triggered it — completed, running, or orphaned, with real durations and a drill-down from session → task → individual agent.
Skill & agent attribution
Exact token share per skill and per subagent type, sourced directly from Claude Code's own telemetry attributes — not inferred or guessed.
Spawn-budget enforcement
An optional hook that actually blocks new agent spawns once your session usage crosses a threshold — not just a dashboard number, a real guardrail wired into Claude Code itself.
Splunk-style time range picker
Every chart supports independent time ranges from 1 minute to 7 days, so you can zoom into a single burst or see the whole week's shape.
Why this exists
This project exists because of one real incident: a session where a burst of agent spawns burned half of a 5-hour usage window in twenty minutes, with no visibility into it happening until the damage was done. Every panel here traces back to actually catching that kind of thing — live, on your own machine, before it costs you the rest of your session.